An advanced compiler designed for a VLIW DSP for sensors-based systems.

Area of Science:

• Computer Architecture
• Compiler Design
• Embedded Systems

Background:

• Very Long Instruction Word (VLIW) architecture offers significant instruction-level parallelism, leading to enhanced computation power and energy efficiency.
• VLIW processor performance is heavily reliant on static compilation, making compiler behavior a critical factor.
• Future sensor-based systems require high performance and energy efficiency, aligning with VLIW capabilities.

Purpose of the Study:

• To develop an advanced compiler for the VLIW Digital Signal Processor (DSP) named Magnolia.
• To optimize the compiler for sensor-based systems by implementing advanced techniques.
• To validate the compiler's effectiveness using standard benchmarks.

Main Methods:

• The compiler development is based on the Open64 compiler infrastructure.
• Advanced optimization techniques were implemented, achieving O3 level optimization.
• The DSPstone test suite was utilized for compiler verification and performance evaluation.

Main Results:

• The developed compiler successfully implemented advanced optimization techniques.
• Performance benchmarks using the DSPstone suite were conducted.
• The generated code demonstrated competitive performance for the Magnolia VLIW DSP.

Conclusions:

• The advanced compiler significantly enhances the performance of the Magnolia VLIW DSP.
• The compiler's capabilities enable the Magnolia processor to meet the demands of future sensor-based systems.
• The optimized Magnolia processor achieves performance on par with current state-of-the-art DSPs.